首页 > 最新文献

arXiv - MATH - Numerical Analysis最新文献

英文 中文
Anisotropic Error Analysis of Weak Galerkin finite element method for Singularly Perturbed Biharmonic Problems 弱 Galerkin 有限元方法对奇异扰动双谐波问题的各向异性误差分析
Pub Date : 2024-09-11 DOI: arxiv-2409.07217
Aayushman Raina, Srinivasan Natesan, Şuayip Toprakseven
We consider the Weak Galerkin finite element approximation of the SingularlyPerturbed Biharmonic elliptic problem on a unit square domain with clampedboundary conditions. Shishkin mesh is used for domain discretization as thesolution exhibits boundary layers near the domain boundary. Error estimates inthe equivalent $H^{2}-$ norm have been established and the uniform convergenceof the proposed method has been proved. Numerical examples are presentedcorroborating our theoretical findings.
我们考虑在一个具有箝位边界条件的单位正方形域上对奇异扰动比谐椭圆问题进行弱 Galerkin 有限元近似。由于解在域边界附近会出现边界层,因此采用 Shishkin 网格进行域离散化。建立了等效 $H^{2}-$ 准则的误差估计,并证明了所提方法的均匀收敛性。给出的数值示例证实了我们的理论发现。
{"title":"Anisotropic Error Analysis of Weak Galerkin finite element method for Singularly Perturbed Biharmonic Problems","authors":"Aayushman Raina, Srinivasan Natesan, Şuayip Toprakseven","doi":"arxiv-2409.07217","DOIUrl":"https://doi.org/arxiv-2409.07217","url":null,"abstract":"We consider the Weak Galerkin finite element approximation of the Singularly\u0000Perturbed Biharmonic elliptic problem on a unit square domain with clamped\u0000boundary conditions. Shishkin mesh is used for domain discretization as the\u0000solution exhibits boundary layers near the domain boundary. Error estimates in\u0000the equivalent $H^{2}-$ norm have been established and the uniform convergence\u0000of the proposed method has been proved. Numerical examples are presented\u0000corroborating our theoretical findings.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"32 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181957","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Dynamic Error-Bounded Hierarchical Matrices in Neural Network Compression 神经网络压缩中的动态误差约束层次矩阵
Pub Date : 2024-09-11 DOI: arxiv-2409.07028
John Mango, Ronald Katende
This paper presents an innovative framework that integrates hierarchicalmatrix (H-matrix) compression techniques into the structure and training ofPhysics-Informed Neural Networks (PINNs). By leveraging the low-rank propertiesof matrix sub-blocks, the proposed dynamic, error-bounded H-matrix compressionmethod significantly reduces computational complexity and storage requirementswithout compromising accuracy. This approach is rigorously compared totraditional compression techniques, such as Singular Value Decomposition (SVD),pruning, and quantization, demonstrating superior performance, particularly inmaintaining the Neural Tangent Kernel (NTK) properties critical for thestability and convergence of neural networks. The findings reveal that H-matrixcompression not only enhances training efficiency but also ensures thescalability and robustness of PINNs for complex, large-scale applications inphysics-based modeling. This work offers a substantial contribution to theoptimization of deep learning models, paving the way for more efficient andpractical implementations of PINNs in real-world scenarios.
本文提出了一种创新框架,将分层矩阵(H-matrix)压缩技术集成到物理信息神经网络(PINNs)的结构和训练中。通过利用矩阵子块的低秩特性,所提出的动态、有误差限制的 H 矩阵压缩方法大大降低了计算复杂度和存储要求,同时不影响准确性。该方法与奇异值分解(SVD)、剪枝和量化等传统压缩技术进行了严格比较,显示出卓越的性能,尤其是在保持对神经网络的稳定性和收敛性至关重要的神经切分核(NTK)特性方面。研究结果表明,H-matrix 压缩不仅能提高训练效率,还能确保 PINNs 的可扩展性和鲁棒性,适用于基于物理学建模的复杂、大规模应用。这项工作为深度学习模型的优化做出了重大贡献,为 PINN 在现实世界场景中更高效、更实用的实现铺平了道路。
{"title":"Dynamic Error-Bounded Hierarchical Matrices in Neural Network Compression","authors":"John Mango, Ronald Katende","doi":"arxiv-2409.07028","DOIUrl":"https://doi.org/arxiv-2409.07028","url":null,"abstract":"This paper presents an innovative framework that integrates hierarchical\u0000matrix (H-matrix) compression techniques into the structure and training of\u0000Physics-Informed Neural Networks (PINNs). By leveraging the low-rank properties\u0000of matrix sub-blocks, the proposed dynamic, error-bounded H-matrix compression\u0000method significantly reduces computational complexity and storage requirements\u0000without compromising accuracy. This approach is rigorously compared to\u0000traditional compression techniques, such as Singular Value Decomposition (SVD),\u0000pruning, and quantization, demonstrating superior performance, particularly in\u0000maintaining the Neural Tangent Kernel (NTK) properties critical for the\u0000stability and convergence of neural networks. The findings reveal that H-matrix\u0000compression not only enhances training efficiency but also ensures the\u0000scalability and robustness of PINNs for complex, large-scale applications in\u0000physics-based modeling. This work offers a substantial contribution to the\u0000optimization of deep learning models, paving the way for more efficient and\u0000practical implementations of PINNs in real-world scenarios.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"10 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181964","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Finite element approximation of stationary Fokker--Planck--Kolmogorov equations with application to periodic numerical homogenization 静态福克--普朗克--科尔莫戈罗夫方程的有限元近似与周期数值同质化的应用
Pub Date : 2024-09-11 DOI: arxiv-2409.07371
Timo Sprekeler, Endre Süli, Zhiwen Zhang
We propose and rigorously analyze a finite element method for theapproximation of stationary Fokker--Planck--Kolmogorov (FPK) equations subjectto periodic boundary conditions in two settings: one with weakly differentiablecoefficients, and one with merely essentially bounded measurable coefficientsunder a Cordes-type condition. These problems arise as governing equations forthe invariant measure in the homogenization of nondivergence-form equationswith large drifts. In particular, the Cordes setting guarantees the existenceand uniqueness of a square-integrable invariant measure. We then suggest andrigorously analyze an approximation scheme for the effective diffusion matrixin both settings, based on the finite element scheme for stationary FPKproblems developed in the first part. Finally, we demonstrate the performanceof the methods through numerical experiments.
我们提出并严格分析了一种有限元方法,用于在两种情况下逼近受周期性边界条件限制的静态福克-普朗克-科尔莫哥罗夫(FPK)方程:一种是系数弱可微的;另一种是在科尔德斯类型条件下系数仅为基本有界可测的。这些问题是在具有大漂移的非发散形式方程的均质化中作为不变量的支配方程出现的。特别是,Cordes 设置保证了平方可积分不变量的存在性和唯一性。然后,我们以第一部分中开发的静态 FPK 问题有限元方案为基础,提出并认真分析了这两种情况下有效扩散矩阵的近似方案。最后,我们通过数值实验证明了这些方法的性能。
{"title":"Finite element approximation of stationary Fokker--Planck--Kolmogorov equations with application to periodic numerical homogenization","authors":"Timo Sprekeler, Endre Süli, Zhiwen Zhang","doi":"arxiv-2409.07371","DOIUrl":"https://doi.org/arxiv-2409.07371","url":null,"abstract":"We propose and rigorously analyze a finite element method for the\u0000approximation of stationary Fokker--Planck--Kolmogorov (FPK) equations subject\u0000to periodic boundary conditions in two settings: one with weakly differentiable\u0000coefficients, and one with merely essentially bounded measurable coefficients\u0000under a Cordes-type condition. These problems arise as governing equations for\u0000the invariant measure in the homogenization of nondivergence-form equations\u0000with large drifts. In particular, the Cordes setting guarantees the existence\u0000and uniqueness of a square-integrable invariant measure. We then suggest and\u0000rigorously analyze an approximation scheme for the effective diffusion matrix\u0000in both settings, based on the finite element scheme for stationary FPK\u0000problems developed in the first part. Finally, we demonstrate the performance\u0000of the methods through numerical experiments.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"8 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181928","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Dual scale Residual-Network for turbulent flow sub grid scale resolving: A prior analysis 用于湍流子网格尺度解析的双尺度残差网络:先验分析
Pub Date : 2024-09-11 DOI: arxiv-2409.07605
Omar Sallam, Mirjam Fürth
This paper introduces generative Residual Networks (ResNet) as a surrogateMachine Learning (ML) tool for Large Eddy Simulation (LES) Sub Grid Scale (SGS)resolving. The study investigates the impact of incorporating Dual ScaleResidual Blocks (DS-RB) within the ResNet architecture. Two LES SGS resolvingmodels are proposed and tested for prior analysis test cases: asuper-resolution model (SR-ResNet) and a SGS stress tensor inference model(SGS-ResNet). The SR-ResNet model task is to upscale LES solutions from coarseto finer grids by inferring unresolved SGS velocity fluctuations, exhibitingsuccess in preserving high-frequency velocity fluctuation information, andaligning with higher-resolution LES solutions' energy spectrum. Furthermore,employing DS-RB enhances prediction accuracy and precision of high-frequencyvelocity fields compared to Single Scale Residual Blocks (SS-RB), evident inboth spatial and spectral domains. The SR-ResNet model is tested and trained onfiltered/downsampled 2-D LES planar jet injection problems at two Reynoldsnumbers, two jet configurations, and two upscale ratios. In the case of SGSstress tensor inference, both SS-RB and DS-RB exhibit higher predictionaccuracy over the Smagorinsky model with reference to the true DNS SGS stresstensor, with DS-RB-based SGS-ResNet showing stronger statistical alignment withDNS data. The SGS-ResNet model is tested on a filtered/downsampled 2-D DNSisotropic homogenous decay turbulence problem. The adoption of DS-RB incursnotable increases in network size, training time, and forward inference time,with the network size expanding by over tenfold, and training and forwardinference times increasing by approximately 0.5 and 3 times, respectively.
本文介绍了作为大涡模拟(LES)子网格尺度(SGS)解析的替代机器学习(ML)工具的生成残差网络(ResNet)。该研究探讨了在 ResNet 架构中加入双尺度残差块(DS-RB)的影响。针对先期分析测试案例,提出并测试了两种 LES SGS 解析模型:超解析模型(SR-ResNet)和 SGS 应力张量推理模型(SGS-ResNet)。SR-ResNet 模型的任务是通过推断未解决的 SGS 速度波动,将 LES 解决方案从粗网格提升到更精细的网格,成功保留了高频速度波动信息,并与更高分辨率 LES 解决方案的能谱相一致。此外,与单尺度残差块(SS-RB)相比,采用 DS-RB 提高了高频速度场的预测精度和准确性,这在空间和频谱域都很明显。SR-ResNet 模型在两种雷诺数、两种喷流配置和两种升尺度比的过滤/降采样 2-D LES 平面喷流注入问题上进行了测试和训练。在 SGS 应力张量推断方面,参照真实 DNS SGS 应变张量,SS-RB 和 DS-RB 都比 Smagorinsky 模型显示出更高的预测精度,基于 DS-RB 的 SGS-ResNet 与 DNS 数据显示出更强的统计一致性。SGS-ResNet 模型在滤波/降采样的 2-D DNS 各向同性均质衰减湍流问题上进行了测试。采用 DS-RB 后,网络规模、训练时间和前向推理时间显著增加,网络规模扩大了 10 倍以上,训练时间和前向推理时间分别增加了约 0.5 倍和 3 倍。
{"title":"Dual scale Residual-Network for turbulent flow sub grid scale resolving: A prior analysis","authors":"Omar Sallam, Mirjam Fürth","doi":"arxiv-2409.07605","DOIUrl":"https://doi.org/arxiv-2409.07605","url":null,"abstract":"This paper introduces generative Residual Networks (ResNet) as a surrogate\u0000Machine Learning (ML) tool for Large Eddy Simulation (LES) Sub Grid Scale (SGS)\u0000resolving. The study investigates the impact of incorporating Dual Scale\u0000Residual Blocks (DS-RB) within the ResNet architecture. Two LES SGS resolving\u0000models are proposed and tested for prior analysis test cases: a\u0000super-resolution model (SR-ResNet) and a SGS stress tensor inference model\u0000(SGS-ResNet). The SR-ResNet model task is to upscale LES solutions from coarse\u0000to finer grids by inferring unresolved SGS velocity fluctuations, exhibiting\u0000success in preserving high-frequency velocity fluctuation information, and\u0000aligning with higher-resolution LES solutions' energy spectrum. Furthermore,\u0000employing DS-RB enhances prediction accuracy and precision of high-frequency\u0000velocity fields compared to Single Scale Residual Blocks (SS-RB), evident in\u0000both spatial and spectral domains. The SR-ResNet model is tested and trained on\u0000filtered/downsampled 2-D LES planar jet injection problems at two Reynolds\u0000numbers, two jet configurations, and two upscale ratios. In the case of SGS\u0000stress tensor inference, both SS-RB and DS-RB exhibit higher prediction\u0000accuracy over the Smagorinsky model with reference to the true DNS SGS stress\u0000tensor, with DS-RB-based SGS-ResNet showing stronger statistical alignment with\u0000DNS data. The SGS-ResNet model is tested on a filtered/downsampled 2-D DNS\u0000isotropic homogenous decay turbulence problem. The adoption of DS-RB incurs\u0000notable increases in network size, training time, and forward inference time,\u0000with the network size expanding by over tenfold, and training and forward\u0000inference times increasing by approximately 0.5 and 3 times, respectively.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"17 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181925","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
A construction of canonical nonconforming finite element spaces for elliptic equations of any order in any dimension 构建任意维度任意阶椭圆方程的典型非符合有限元空间
Pub Date : 2024-09-10 DOI: arxiv-2409.06134
Jia Li, Shuonan Wu
A unified construction of canonical $H^m$-nonconforming finite elements isdeveloped for $n$-dimensional simplices for any $m, n geq 1$. Consistency withthe Morley-Wang-Xu elements [Math. Comp. 82 (2013), pp. 25-43] is maintainedwhen $m leq n$. In the general case, the degrees of freedom and the shapefunction space exhibit well-matched multi-layer structures that ensure theiralignment. Building on the concept of the nonconforming bubble function, theunisolvence is established using an equivalent integral-type representation ofthe shape function space and by applying induction on $m$. The correspondingnonconforming finite element method applies to $2m$-th order elliptic problems,with numerical results for $m=3$ and $m=4$ in 2D supporting the theoreticalanalysis.
针对任意$m, n geq 1$的$n$维单纯形,建立了一个统一的典型$H^m$-不符合有限元的构造。当 $m leq n$ 时,与 Morley-Wang-Xu 元 [Math. Comp. 82 (2013), pp.在一般情况下,自由度和形状函数空间表现出良好匹配的多层结构,确保了它们的对齐。在不符合气泡函数概念的基础上,使用形状函数空间的等效积分型表示,并通过对 $m$ 的归纳,建立了不符合气泡函数。相应的不符合有限元法适用于 2m$-th阶椭圆问题,二维中$m=3$和$m=4$的数值结果支持理论分析。
{"title":"A construction of canonical nonconforming finite element spaces for elliptic equations of any order in any dimension","authors":"Jia Li, Shuonan Wu","doi":"arxiv-2409.06134","DOIUrl":"https://doi.org/arxiv-2409.06134","url":null,"abstract":"A unified construction of canonical $H^m$-nonconforming finite elements is\u0000developed for $n$-dimensional simplices for any $m, n geq 1$. Consistency with\u0000the Morley-Wang-Xu elements [Math. Comp. 82 (2013), pp. 25-43] is maintained\u0000when $m leq n$. In the general case, the degrees of freedom and the shape\u0000function space exhibit well-matched multi-layer structures that ensure their\u0000alignment. Building on the concept of the nonconforming bubble function, the\u0000unisolvence is established using an equivalent integral-type representation of\u0000the shape function space and by applying induction on $m$. The corresponding\u0000nonconforming finite element method applies to $2m$-th order elliptic problems,\u0000with numerical results for $m=3$ and $m=4$ in 2D supporting the theoretical\u0000analysis.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"11 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181969","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Distributed Cooperative AI for Large-Scale Eigenvalue Computations Using Neural Networks 利用神经网络进行大规模特征值计算的分布式合作人工智能
Pub Date : 2024-09-10 DOI: arxiv-2409.06746
Ronald Katende
This paper presents a novel method for eigenvalue computation using adistributed cooperative neural network framework. Unlike traditional techniquesthat struggle with scalability in large systems, our decentralized algorithmenables multiple autonomous agents to collaboratively estimate the smallesteigenvalue of large matrices. Each agent uses a localized neural network model,refining its estimates through inter-agent communication. Our approachguarantees convergence to the true eigenvalue, even with communication failuresor network disruptions. Theoretical analysis confirms the robustness andaccuracy of the method, while empirical results demonstrate its betterperformance compared to some traditional centralized algorithms
本文提出了一种利用分布式合作神经网络框架进行特征值计算的新方法。与在大型系统中难以扩展的传统技术不同,我们的去中心化算法使多个自主代理能够协同估计大型矩阵的小特征值。每个代理使用一个本地化神经网络模型,通过代理间通信完善其估计值。我们的方法能保证收敛到真正的特征值,即使在通信失败或网络中断的情况下也是如此。理论分析证实了该方法的稳健性和准确性,而实证结果表明,与一些传统的集中式算法相比,它的性能更好
{"title":"Distributed Cooperative AI for Large-Scale Eigenvalue Computations Using Neural Networks","authors":"Ronald Katende","doi":"arxiv-2409.06746","DOIUrl":"https://doi.org/arxiv-2409.06746","url":null,"abstract":"This paper presents a novel method for eigenvalue computation using a\u0000distributed cooperative neural network framework. Unlike traditional techniques\u0000that struggle with scalability in large systems, our decentralized algorithm\u0000enables multiple autonomous agents to collaboratively estimate the smallest\u0000eigenvalue of large matrices. Each agent uses a localized neural network model,\u0000refining its estimates through inter-agent communication. Our approach\u0000guarantees convergence to the true eigenvalue, even with communication failures\u0000or network disruptions. Theoretical analysis confirms the robustness and\u0000accuracy of the method, while empirical results demonstrate its better\u0000performance compared to some traditional centralized algorithms","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"50 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181962","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
A tutorial on automatic differentiation with complex numbers 复数自动微分教程
Pub Date : 2024-09-10 DOI: arxiv-2409.06752
Nicholas Krämer
Automatic differentiation is everywhere, but there exists only minimaldocumentation of how it works in complex arithmetic beyond stating "derivativesin $mathbb{C}^d$" $cong$ "derivatives in $mathbb{R}^{2d}$" and, at best,shallow references to Wirtinger calculus. Unfortunately, the equivalence$mathbb{C}^d cong mathbb{R}^{2d}$ becomes insufficient as soon as we need toderive custom gradient rules, e.g., to avoid differentiating "through"expensive linear algebra functions or differential equation simulators. Tocombat such a lack of documentation, this article surveys forward- andreverse-mode automatic differentiation with complex numbers, covering topicssuch as Wirtinger derivatives, a modified chain rule, and different gradientconventions while explicitly avoiding holomorphicity and the Cauchy--Riemannequations (which would be far too restrictive). To be precise, we will derive,explain, and implement a complex version of Jacobian-vector and vector-Jacobianproducts almost entirely with linear algebra without relying on complexanalysis or differential geometry. This tutorial is a call to action, for usersand developers alike, to take complex values seriously when implementing customgradient propagation rules -- the manuscript explains how.
自动微分无处不在,但关于它如何在复杂算术中起作用,除了说明"$mathbb{C}^d$中的导数"$cong$"$mathbb{R}^{2d}$中的导数 "以及充其量浅显地引用维廷格微积分之外,只有极少的文档。不幸的是,一旦我们需要导出自定义梯度规则,例如,为了避免 "通过 "昂贵的线性代数函数或微分方程模拟器进行微分,等价$mathbb{C}^d cong mathbb{R}^{2d}$就变得不够了。为了解决文献缺乏的问题,本文研究了复数的正向和反向自动微分,涵盖了 Wirtinger 导数、修正的链式规则和不同的梯度约定等主题,同时明确避免了全形性和 Cauchy--Rieman 问题(限制性太大)。准确地说,我们将推导、解释和实现复杂版本的雅各布向量和向量-雅各布积,几乎完全使用线性代数,而不依赖复杂分析或微分几何。本教程呼吁用户和开发人员在实施自定义梯度传播规则时,认真对待复杂值--手稿将解释如何实现。
{"title":"A tutorial on automatic differentiation with complex numbers","authors":"Nicholas Krämer","doi":"arxiv-2409.06752","DOIUrl":"https://doi.org/arxiv-2409.06752","url":null,"abstract":"Automatic differentiation is everywhere, but there exists only minimal\u0000documentation of how it works in complex arithmetic beyond stating \"derivatives\u0000in $mathbb{C}^d$\" $cong$ \"derivatives in $mathbb{R}^{2d}$\" and, at best,\u0000shallow references to Wirtinger calculus. Unfortunately, the equivalence\u0000$mathbb{C}^d cong mathbb{R}^{2d}$ becomes insufficient as soon as we need to\u0000derive custom gradient rules, e.g., to avoid differentiating \"through\"\u0000expensive linear algebra functions or differential equation simulators. To\u0000combat such a lack of documentation, this article surveys forward- and\u0000reverse-mode automatic differentiation with complex numbers, covering topics\u0000such as Wirtinger derivatives, a modified chain rule, and different gradient\u0000conventions while explicitly avoiding holomorphicity and the Cauchy--Riemann\u0000equations (which would be far too restrictive). To be precise, we will derive,\u0000explain, and implement a complex version of Jacobian-vector and vector-Jacobian\u0000products almost entirely with linear algebra without relying on complex\u0000analysis or differential geometry. This tutorial is a call to action, for users\u0000and developers alike, to take complex values seriously when implementing custom\u0000gradient propagation rules -- the manuscript explains how.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"171 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181961","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Randomized low-rank Runge-Kutta methods 随机低阶 Runge-Kutta 方法
Pub Date : 2024-09-10 DOI: arxiv-2409.06384
Hei Yin Lam, Gianluca Ceruti, Daniel Kressner
This work proposes and analyzes a new class of numerical integrators forcomputing low-rank approximations to solutions of matrix differential equation.We combine an explicit Runge-Kutta method with repeated randomized low-rankapproximation to keep the rank of the stages limited. The so-called generalizedNystr"om method is particularly well suited for this purpose; it buildslow-rank approximations from random sketches of the discretized dynamics. Incontrast, all existing dynamical low-rank approximation methods aredeterministic and usually perform tangent space projections to limit rankgrowth. Using such tangential projections can result in larger error comparedto approximating the dynamics directly. Moreover, sketching allows forincreased flexibility and efficiency by choosing structured random matricesadapted to the structure of the matrix differential equation. Under suitableassumptions, we establish moment and tail bounds on the error of our randomizedlow-rank Runge-Kutta methods. When combining the classical Runge-Kutta methodwith generalized Nystr"om, we obtain a method called Rand RK4, which exhibitsfourth-order convergence numerically -- up to the low-rank approximation error.For a modified variant of Rand RK4, we also establish fourth-order convergencetheoretically. Numerical experiments for a range of examples from theliterature demonstrate that randomized low-rank Runge-Kutta methods comparefavorably with two popular dynamical low-rank approximation methods, in termsof robustness and speed of convergence.
我们将显式 Runge-Kutta 方法与重复随机低秩近似法相结合,以限制各阶段的秩。所谓的广义 Nystr"om 方法特别适合这一目的;它从离散动力学的随机草图中建立低阶近似。相比之下,所有现有的动力学低阶近似方法都是确定性的,通常执行切向空间投影来限制阶数增长。与直接逼近动力学相比,使用这种切向投影会导致更大的误差。此外,草图法通过选择与矩阵微分方程结构相适应的结构化随机矩阵,提高了灵活性和效率。在合适的假设条件下,我们建立了随机低秩 Runge-Kutta 方法的误差矩界和尾界。当把经典 Runge-Kutta 方法与广义 Nystr"om 结合起来时,我们得到了一种称为 Rand RK4 的方法,它在数值上表现出四阶收敛性--直到低阶近似误差为止。对文献中一系列例子的数值实验表明,随机低阶 Runge-Kutta 方法与两种流行的动态低阶近似方法相比,在鲁棒性和收敛速度方面都更胜一筹。
{"title":"Randomized low-rank Runge-Kutta methods","authors":"Hei Yin Lam, Gianluca Ceruti, Daniel Kressner","doi":"arxiv-2409.06384","DOIUrl":"https://doi.org/arxiv-2409.06384","url":null,"abstract":"This work proposes and analyzes a new class of numerical integrators for\u0000computing low-rank approximations to solutions of matrix differential equation.\u0000We combine an explicit Runge-Kutta method with repeated randomized low-rank\u0000approximation to keep the rank of the stages limited. The so-called generalized\u0000Nystr\"om method is particularly well suited for this purpose; it builds\u0000low-rank approximations from random sketches of the discretized dynamics. In\u0000contrast, all existing dynamical low-rank approximation methods are\u0000deterministic and usually perform tangent space projections to limit rank\u0000growth. Using such tangential projections can result in larger error compared\u0000to approximating the dynamics directly. Moreover, sketching allows for\u0000increased flexibility and efficiency by choosing structured random matrices\u0000adapted to the structure of the matrix differential equation. Under suitable\u0000assumptions, we establish moment and tail bounds on the error of our randomized\u0000low-rank Runge-Kutta methods. When combining the classical Runge-Kutta method\u0000with generalized Nystr\"om, we obtain a method called Rand RK4, which exhibits\u0000fourth-order convergence numerically -- up to the low-rank approximation error.\u0000For a modified variant of Rand RK4, we also establish fourth-order convergence\u0000theoretically. Numerical experiments for a range of examples from the\u0000literature demonstrate that randomized low-rank Runge-Kutta methods compare\u0000favorably with two popular dynamical low-rank approximation methods, in terms\u0000of robustness and speed of convergence.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"55 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181966","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
An Eulerian Vortex Method on Flow Maps 流图上的欧拉旋涡法
Pub Date : 2024-09-10 DOI: arxiv-2409.06201
Sinan Wang, Yitong Deng, Molin Deng, Hong-Xing Yu, Junwei Zhou, Duowen Chen, Taku Komura, Jiajun Wu, Bo Zhu
We present an Eulerian vortex method based on the theory of flow maps tosimulate the complex vortical motions of incompressible fluids. Central to ourmethod is the novel incorporation of the flow-map transport equations for lineelements, which, in combination with a bi-directional marching scheme for flowmaps, enables the high-fidelity Eulerian advection of vorticity variables. Thefundamental motivation is that, compared to impulse $mathbf{m}$, which hasbeen recently bridged with flow maps to encouraging results, vorticity$boldsymbol{omega}$ promises to be preferable for its numerically stabilityand physical interpretability. To realize the full potential of this novelformulation, we develop a new Poisson solving scheme for vorticity-to-velocityreconstruction that is both efficient and able to accurately handle thecoupling near solid boundaries. We demonstrate the efficacy of our approachwith a range of vortex simulation examples, including leapfrog vortices, vortexcollisions, cavity flow, and the formation of complex vortical structures dueto solid-fluid interactions.
我们提出了一种基于流图理论的欧拉旋涡方法,用于模拟不可压缩流体的复杂旋涡运动。我们方法的核心是新颖地加入了线元的流图传输方程,结合流图的双向行进方案,实现了涡度变量的高保真欧拉平流。其根本原因在于,与脉冲$mathbf{m}$相比,涡度$boldsymbol{omega}$在数值稳定性和物理可解释性方面更胜一筹。为了充分发挥这种新颖形式的潜力,我们开发了一种新的泊松求解方案,用于涡度-速度重建,既高效又能准确处理固体边界附近的耦合。我们用一系列涡旋模拟实例证明了我们方法的有效性,包括跃迁涡旋、涡旋碰撞、空腔流以及固液相互作用形成的复杂涡旋结构。
{"title":"An Eulerian Vortex Method on Flow Maps","authors":"Sinan Wang, Yitong Deng, Molin Deng, Hong-Xing Yu, Junwei Zhou, Duowen Chen, Taku Komura, Jiajun Wu, Bo Zhu","doi":"arxiv-2409.06201","DOIUrl":"https://doi.org/arxiv-2409.06201","url":null,"abstract":"We present an Eulerian vortex method based on the theory of flow maps to\u0000simulate the complex vortical motions of incompressible fluids. Central to our\u0000method is the novel incorporation of the flow-map transport equations for line\u0000elements, which, in combination with a bi-directional marching scheme for flow\u0000maps, enables the high-fidelity Eulerian advection of vorticity variables. The\u0000fundamental motivation is that, compared to impulse $mathbf{m}$, which has\u0000been recently bridged with flow maps to encouraging results, vorticity\u0000$boldsymbol{omega}$ promises to be preferable for its numerically stability\u0000and physical interpretability. To realize the full potential of this novel\u0000formulation, we develop a new Poisson solving scheme for vorticity-to-velocity\u0000reconstruction that is both efficient and able to accurately handle the\u0000coupling near solid boundaries. We demonstrate the efficacy of our approach\u0000with a range of vortex simulation examples, including leapfrog vortices, vortex\u0000collisions, cavity flow, and the formation of complex vortical structures due\u0000to solid-fluid interactions.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"118 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181972","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
A Liang-Kleeman Causality Analysis based on Linear Inverse Modeling 基于线性反演模型的梁-克莱曼因果关系分析
Pub Date : 2024-09-10 DOI: arxiv-2409.06797
Justin Lien
Causality analysis is a powerful tool for determining cause-and-effectrelationships between variables in a system by quantifying the influence of onevariable on another. Despite significant advancements in the field, manyexisting studies are constrained by their focus on unidirectional causality orGaussian external forcing, limiting their applicability to complex real-worldproblems. This study proposes a novel data-driven approach to causalityanalysis for complex stochastic differential systems, integrating the conceptsof Liang-Kleeman information flow and linear inverse modeling. Our methodmodels environmental noise as either memoryless Gaussian white noise ormemory-retaining Ornstein-Uhlenbeck colored noise, and allows for self andmutual causality, providing a more realistic representation and interpretationof the underlying system. Moreover, this LIM-based approach can identify theindividual contribution of dynamics and correlation to causality. We apply thisapproach to re-examine the causal relationships between the ElNi~{n}o-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD), twomajor climate phenomena that significantly influence global climate patterns.In general, regardless of the type of noise used, the causality between ENSOand IOD is mutual but asymmetric, with the causality map reflecting anENSO-like pattern consistent with previous studies. Notably, in the case ofcolored noise, the noise memory map reveals a hotspot in the Ni~no 3 region,which is further related to the information flow. This suggests that ourapproach offers a more comprehensive framework and provides deeper insightsinto the causal inference of global climate systems.
因果分析是通过量化一个变量对另一个变量的影响来确定系统中变量之间因果关系的有力工具。尽管该领域取得了重大进展,但现有的许多研究都局限于单向因果关系或高斯外力作用,从而限制了它们对复杂现实问题的适用性。本研究针对复杂随机微分系统的因果关系分析,提出了一种新颖的数据驱动方法,整合了梁-克莱曼信息流和线性逆建模的概念。我们的方法将环境噪声建模为无记忆高斯白噪声或有记忆奥恩斯坦-乌伦贝克彩色噪声,并允许自因和互因,从而提供了对底层系统更真实的表示和解释。此外,这种基于 LIM 的方法可以识别动态和相关性对因果关系的单独贡献。一般来说,无论使用哪种噪声,厄尔尼诺/南方涛动和印度洋偶极子之间的因果关系都是相互的,但并不对称,因果关系图反映了类似厄尔尼诺/南方涛动的模式,这与之前的研究一致。值得注意的是,在彩色噪声的情况下,噪声记忆图显示了 Ni~no 3 区域的热点,这与信息流进一步相关。这表明我们的方法提供了一个更全面的框架,为全球气候系统的因果推断提供了更深刻的见解。
{"title":"A Liang-Kleeman Causality Analysis based on Linear Inverse Modeling","authors":"Justin Lien","doi":"arxiv-2409.06797","DOIUrl":"https://doi.org/arxiv-2409.06797","url":null,"abstract":"Causality analysis is a powerful tool for determining cause-and-effect\u0000relationships between variables in a system by quantifying the influence of one\u0000variable on another. Despite significant advancements in the field, many\u0000existing studies are constrained by their focus on unidirectional causality or\u0000Gaussian external forcing, limiting their applicability to complex real-world\u0000problems. This study proposes a novel data-driven approach to causality\u0000analysis for complex stochastic differential systems, integrating the concepts\u0000of Liang-Kleeman information flow and linear inverse modeling. Our method\u0000models environmental noise as either memoryless Gaussian white noise or\u0000memory-retaining Ornstein-Uhlenbeck colored noise, and allows for self and\u0000mutual causality, providing a more realistic representation and interpretation\u0000of the underlying system. Moreover, this LIM-based approach can identify the\u0000individual contribution of dynamics and correlation to causality. We apply this\u0000approach to re-examine the causal relationships between the El\u0000Ni~{n}o-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD), two\u0000major climate phenomena that significantly influence global climate patterns.\u0000In general, regardless of the type of noise used, the causality between ENSO\u0000and IOD is mutual but asymmetric, with the causality map reflecting an\u0000ENSO-like pattern consistent with previous studies. Notably, in the case of\u0000colored noise, the noise memory map reveals a hotspot in the Ni~no 3 region,\u0000which is further related to the information flow. This suggests that our\u0000approach offers a more comprehensive framework and provides deeper insights\u0000into the causal inference of global climate systems.","PeriodicalId":501162,"journal":{"name":"arXiv - MATH - Numerical Analysis","volume":"55 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142181960","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
期刊
arXiv - MATH - Numerical Analysis
全部 Acc. Chem. Res. ACS Applied Bio Materials ACS Appl. Electron. Mater. ACS Appl. Energy Mater. ACS Appl. Mater. Interfaces ACS Appl. Nano Mater. ACS Appl. Polym. Mater. ACS BIOMATER-SCI ENG ACS Catal. ACS Cent. Sci. ACS Chem. Biol. ACS Chemical Health & Safety ACS Chem. Neurosci. ACS Comb. Sci. ACS Earth Space Chem. ACS Energy Lett. ACS Infect. Dis. ACS Macro Lett. ACS Mater. Lett. ACS Med. Chem. Lett. ACS Nano ACS Omega ACS Photonics ACS Sens. ACS Sustainable Chem. Eng. ACS Synth. Biol. Anal. Chem. BIOCHEMISTRY-US Bioconjugate Chem. BIOMACROMOLECULES Chem. Res. Toxicol. Chem. Rev. Chem. Mater. CRYST GROWTH DES ENERG FUEL Environ. Sci. Technol. Environ. Sci. Technol. Lett. Eur. J. Inorg. Chem. IND ENG CHEM RES Inorg. Chem. J. Agric. Food. Chem. J. Chem. Eng. Data J. Chem. Educ. J. Chem. Inf. Model. J. Chem. Theory Comput. J. Med. Chem. J. Nat. Prod. J PROTEOME RES J. Am. Chem. Soc. LANGMUIR MACROMOLECULES Mol. Pharmaceutics Nano Lett. Org. Lett. ORG PROCESS RES DEV ORGANOMETALLICS J. Org. Chem. J. Phys. Chem. J. Phys. Chem. A J. Phys. Chem. B J. Phys. Chem. C J. Phys. Chem. Lett. Analyst Anal. Methods Biomater. Sci. Catal. Sci. Technol. Chem. Commun. Chem. Soc. Rev. CHEM EDUC RES PRACT CRYSTENGCOMM Dalton Trans. Energy Environ. Sci. ENVIRON SCI-NANO ENVIRON SCI-PROC IMP ENVIRON SCI-WAT RES Faraday Discuss. Food Funct. Green Chem. Inorg. Chem. Front. Integr. Biol. J. Anal. At. Spectrom. J. Mater. Chem. A J. Mater. Chem. B J. Mater. Chem. C Lab Chip Mater. Chem. Front. Mater. Horiz. MEDCHEMCOMM Metallomics Mol. Biosyst. Mol. Syst. Des. Eng. Nanoscale Nanoscale Horiz. Nat. Prod. Rep. New J. Chem. Org. Biomol. Chem. Org. Chem. Front. PHOTOCH PHOTOBIO SCI PCCP Polym. Chem.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1